A discriminator includes a common learning unit and a plurality of learning units that are connected to an output unit of the common learning unit. The discriminator is trained, using a plurality of data sets of a medical image and an image data of a first disease region, such that information indicating the first disease region is output from a first learning unit in a case in which the medical image is input to the common learning unit. The discriminator is trained, using a plurality of data sets of a medical image and an image data of a second disease region having at least one of a medical causal relationship or an anatomic causal relationship with the first disease, such that information indicating the second disease region is output from a second learning unit in a case in which the medical image is input to the common learning unit.
Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A learning method that trains a discriminator comprising a common convolutional neural network (CNN) that includes an input unit and an output unit and a plurality of CNNs each of which includes an input unit which is connected to the output unit of the common CNN and an output unit, the method comprising: training the common CNN and a first CNN among the plurality of CNNs of the discriminator, using a plurality of data sets of a medical image and an image data of a first disease region in which a first disease appears in the medical image, such that information indicating the first disease region is output from the output unit of the first CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN; and training the common CNN and a second CNN among the plurality of CNNs of the discriminator, using a plurality of data sets of a medical image and an image data of a second disease region in which a second disease having at least one of a medical causal relationship or an anatomic causal relationship with the first disease appears in the medical image, such that information indicating the second disease region is output from the output unit of the second CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN, wherein after the discriminator being trained, the common CNN outputs a feature amount map common to a case in which the first disease is discriminated and a case in which the second disease is discriminated.
2. The learning method according to claim 1 , wherein the common CNN is trained, using the plurality of data sets of a medical image and an image data of a first disease region in which the first disease appears in the medical image and the plurality of data sets of a medical image and an image data of a second disease region in which the second disease appears in the medical image, such that a feature amount data of the medical image is output from the output unit of the common CNN in a case in which the medical image is input to the input unit of the common CNN.
This invention relates to a machine learning method for medical image analysis, specifically for training a convolutional neural network (CNN) to extract feature data from medical images containing disease regions. The method addresses the challenge of accurately identifying and distinguishing between different disease regions in medical images, such as those affected by a first disease and a second disease. The method involves training a common CNN using multiple datasets, each consisting of a medical image and corresponding image data of a disease region where a specific disease (either the first or second disease) appears. The CNN is trained to output feature data from its output unit when a medical image is input to its input unit. This feature data represents the extracted characteristics of the medical image, enabling the identification of disease regions. The training process ensures the CNN can generalize across different disease types, improving diagnostic accuracy and efficiency in medical imaging applications. The method leverages the CNN's ability to learn hierarchical features, making it suitable for tasks like disease detection, segmentation, and classification in medical images.
3. The learning method according to claim 2 , wherein the discriminator is trained, using a plurality of data sets of a medical image and correct information of an anatomic part of the first disease in the medical image, such that information indicating the anatomic part of the first disease is output from the output unit of a third CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN.
This invention relates to a machine learning method for medical image analysis, specifically for identifying anatomical parts affected by a disease. The method addresses the challenge of accurately detecting and localizing disease-specific anatomical regions in medical images, such as X-rays, MRIs, or CT scans, where variations in image quality, patient anatomy, and disease presentation can complicate automated analysis. The method involves training a convolutional neural network (CNN) architecture with multiple CNNs to process medical images and output information about disease-affected anatomical regions. A discriminator component is trained using datasets containing medical images and corresponding annotations of anatomical parts affected by a specific disease. During training, the discriminator learns to identify and output the relevant anatomical regions when a medical image is input into the shared CNN structure. This approach improves the accuracy and reliability of disease localization in medical imaging, aiding in diagnosis and treatment planning. The method leverages a multi-CNN framework where a common CNN processes the input image, and a third CNN within the network outputs the anatomical part information. The discriminator is trained to ensure that the output correctly identifies the disease-affected regions, enhancing the model's ability to generalize across different medical images. This technique is particularly useful in clinical settings where precise disease localization is critical for effective patient care.
4. The learning method according to claim 1 , wherein the discriminator is trained, using a plurality of data sets of a medical image and correct information of an anatomic part of the first disease in the medical image, such that information indicating the anatomic part of the first disease is output from the output unit of a third CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN.
This invention relates to a machine learning method for medical image analysis, specifically for identifying anatomical parts affected by a disease. The method addresses the challenge of accurately detecting and localizing disease-specific anatomical regions in medical images, which is critical for diagnosis and treatment planning. The system uses a neural network architecture with multiple convolutional neural networks (CNNs) to process medical images and output information about disease-affected anatomical parts. A discriminator component is trained using datasets containing medical images and corresponding correct information about the anatomical regions of a first disease. During training, the discriminator learns to identify and output the relevant anatomical parts when a medical image is input into the system. The architecture leverages shared features extracted by a common CNN, enhancing efficiency and accuracy in disease localization. This approach improves the reliability of automated medical image analysis by ensuring precise identification of disease-specific anatomical regions, which is essential for clinical decision-making. The method is particularly useful in scenarios where manual annotation is time-consuming or impractical, providing a scalable solution for medical imaging applications.
5. The learning method according to claim 1 , wherein the discriminator is trained, using a plurality of data sets of a medical image and information indicating whether or not the second disease is present in the medical image, such that the information indicating whether or not the second disease is present is output from the output unit of a fourth CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN.
This invention relates to a machine learning method for medical image analysis, specifically for detecting the presence of a second disease in medical images. The method addresses the challenge of accurately identifying multiple diseases in medical images, where traditional approaches may struggle with distinguishing between different conditions or detecting secondary diseases that may be less prominent or overlapping with primary conditions. The method involves a neural network architecture comprising multiple convolutional neural networks (CNNs) that share a common CNN structure. The discriminator, a component of this architecture, is trained using datasets containing medical images and corresponding labels indicating whether a second disease is present. During training, a medical image is input into the common CNN, and the discriminator outputs a prediction indicating the presence or absence of the second disease. The discriminator is specifically implemented as a fourth CNN within the plurality of CNNs, ensuring that the shared features extracted by the common CNN are effectively utilized for disease detection. The training process optimizes the discriminator to accurately classify the presence of the second disease, leveraging the shared feature extraction capabilities of the common CNN to improve detection performance. This approach enhances the ability to detect secondary diseases in medical images, even when they may be subtle or coexist with primary conditions. The method is particularly useful in medical diagnostics where early or secondary disease detection is critical for patient outcomes.
6. The learning method according to claim 1 , wherein each of the common CNN and the plurality of CNNs is a neural network that comprises an input layer as the input unit, a plurality of intermediate layers, and an output layer as the output unit.
This invention relates to a neural network-based learning method for improving the performance of convolutional neural networks (CNNs). The method addresses the challenge of efficiently training multiple CNNs while leveraging shared features to enhance accuracy and reduce computational overhead. The system includes a common CNN and a plurality of specialized CNNs, each structured with an input layer, multiple intermediate layers, and an output layer. The common CNN processes input data to extract shared features, which are then passed to the specialized CNNs for further processing. Each specialized CNN refines these features to produce task-specific outputs. The intermediate layers in both the common and specialized CNNs perform convolutional operations, pooling, and activation functions to progressively extract and transform features. The output layer generates final predictions or classifications based on the refined features. This hierarchical approach allows the system to share computational resources while maintaining high accuracy across multiple tasks. The method is particularly useful in applications requiring multi-task learning, such as autonomous systems, medical imaging, and natural language processing, where efficient feature extraction and task-specific adaptation are critical.
7. The learning method according to claim 1 , wherein the first disease is thrombus and the second disease is infarction.
This invention relates to a machine learning method for distinguishing between thrombus and infarction, two distinct cardiovascular conditions. The method involves training a machine learning model to classify medical data, such as imaging or physiological measurements, into one of these two disease categories. The model is designed to analyze input data and output a prediction indicating whether the condition is thrombus or infarction. Thrombus refers to the formation of a blood clot within a blood vessel, while infarction refers to tissue death due to insufficient blood supply, often caused by a blocked artery. The method addresses the challenge of accurately differentiating between these conditions, which is critical for proper diagnosis and treatment. The model may use various types of input data, including but not limited to medical images, sensor readings, or patient records, to improve diagnostic accuracy. The invention aims to enhance early detection and intervention by providing a reliable automated classification system for these cardiovascular diseases.
8. The learning method according to claim 1 , wherein the medical image is a brain image.
This invention relates to a machine learning method for analyzing medical images, specifically brain images, to improve diagnostic accuracy. The method involves training a neural network model using a dataset of brain images to detect and classify abnormalities such as tumors, lesions, or structural anomalies. The neural network is trained to extract relevant features from the brain images, such as tissue density, shape, and texture, to distinguish between healthy and pathological regions. The method may also incorporate additional data, such as patient metadata or clinical history, to enhance prediction accuracy. The trained model can then be applied to new brain images to assist radiologists in diagnosis, reducing human error and improving efficiency. The invention addresses the challenge of accurately interpreting complex brain imaging data, which is critical for early detection and treatment of neurological disorders. The neural network is optimized for high precision and recall, ensuring reliable performance in clinical settings. The method may also include techniques for handling image variability, such as noise reduction and normalization, to improve robustness across different imaging modalities. By automating part of the diagnostic process, the invention aims to reduce the workload on medical professionals while maintaining high diagnostic standards.
9. A discriminator that is trained by the learning method according to claim 1 .
A discriminator is a neural network component used in generative adversarial networks (GANs) to distinguish between real and generated data. The discriminator is trained using a learning method that involves iterative updates to its parameters based on feedback from both real and generated data samples. During training, the discriminator receives real data from a dataset and generated data from a generator network. It outputs a probability score indicating whether the input data is real or fake. The discriminator's parameters are adjusted to minimize the difference between its predictions and the true labels, improving its ability to accurately classify real and generated data. The generator, in turn, is trained to produce data that fools the discriminator, creating a competitive learning process. This adversarial training improves the generator's ability to produce high-quality, realistic data. The discriminator's training involves optimizing a loss function that measures its classification performance, ensuring it becomes increasingly effective at distinguishing real from generated data. This method enhances the overall performance of the GAN by refining both the discriminator and the generator through iterative feedback.
10. A learning apparatus that trains a discriminator comprising a common CNN that includes an input unit and an output unit and a plurality of CNNs each of which includes an input unit which is connected to the output unit of the common CNN and an output unit, the learning apparatus configured to: train the common CNN and a first CNN among the plurality of CNNs of the discriminator, using a plurality of data sets of a medical image and an image data of a first disease region in which a first disease appears in the medical image, such that information indicating the first disease region is output from the output unit of the first CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN; and train the common CNN and a second CNN among the plurality of CNNs of the discriminator, using a plurality of data sets of a medical image and an image data of a second disease region in which a second disease having at least one of a medical causal relationship or an anatomic causal relationship with the first disease appears in the medical image, such that information indicating the second disease region is output from the output unit of the second CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN, wherein after the discriminator being trained, the common CNN outputs a feature amount map common to a case in which the first disease is discriminated and a case in which the second disease is discriminated.
The invention relates to a learning apparatus for training a discriminator in medical image analysis. The discriminator comprises a common convolutional neural network (CNN) with an input and output unit, and multiple CNNs, each connected to the common CNN's output. Each of these CNNs also has an input and output unit. The apparatus trains the common CNN and a first CNN using medical images and annotated data of a first disease region, enabling the first CNN to output information indicating the first disease region when a medical image is input. Similarly, the apparatus trains the common CNN and a second CNN using medical images and annotated data of a second disease region, where the second disease has a medical or anatomical causal relationship with the first disease. The second CNN outputs information indicating the second disease region when a medical image is input. After training, the common CNN generates a feature map that is shared between the first and second disease discrimination tasks. This approach allows for efficient multi-disease detection by leveraging shared features while maintaining task-specific outputs.
11. The learning apparatus according to claim 10 , further configured to: train the common CNN, using the plurality of data sets of a medical image and an image data of a first disease region in which the first disease appears in the medical image and the plurality of data sets of a medical image and an image data of a second disease region in which the second disease appears in the medical image, such that a feature amount data of the medical image is output from the output unit of the common CNN in a case in which the medical image is input to the input unit of the common CNN.
This invention relates to a learning apparatus for training a convolutional neural network (CNN) to analyze medical images for disease detection. The apparatus addresses the challenge of accurately identifying disease regions in medical images by leveraging a shared CNN architecture trained on multiple disease types. The system processes medical images and corresponding disease region data for at least two distinct diseases, enabling the CNN to extract feature data from input medical images. The apparatus includes an input unit for receiving medical images, an output unit for providing feature data, and a training mechanism that uses datasets containing medical images and annotated disease regions for both a first and a second disease. During training, the CNN learns to output feature data that distinguishes between healthy and diseased regions across different conditions. This approach improves diagnostic accuracy by generalizing feature extraction across multiple disease types, reducing the need for separate models for each condition. The apparatus is designed to enhance medical imaging analysis by providing a unified framework for detecting and characterizing diverse pathological regions in medical images.
12. A discriminator that is trained by the learning apparatus according to claim 10 .
A discriminator is a neural network component used in generative adversarial networks (GANs) to distinguish between real and artificially generated data. The discriminator is trained by a learning apparatus that employs a training method involving a generator and the discriminator. The generator creates synthetic data, while the discriminator evaluates the authenticity of the data. The learning apparatus adjusts the discriminator's parameters based on feedback from its performance in distinguishing real data from generated data. The training process involves iterative updates to the discriminator to improve its ability to accurately classify data as real or fake. The discriminator's training is optimized to enhance its discriminative power, ensuring it can effectively differentiate between real and synthetic data. This training method is part of a broader system for improving the quality of generated data by refining the discriminator's ability to detect inconsistencies or artifacts in synthetic outputs. The discriminator's performance is critical in adversarial training, as it directly influences the generator's ability to produce high-quality, realistic data. The learning apparatus may use techniques such as gradient-based optimization to fine-tune the discriminator's parameters during training. The discriminator's architecture may include multiple layers, such as convolutional layers, to process and analyze input data effectively. The training process may also incorporate techniques like label smoothing or adversarial regularization to improve generalization and robustness. The discriminator's training is designed to be scalable and adaptable to different types of data, including images, text, or audio, depending on the application. The overall goal is to develop a d
13. A non-transitory computer readable medium for storing a learning program that trains a discriminator comprising a common CNN that includes an input unit and an output unit and a plurality of CNNs each of which includes an input unit which is connected to the output unit of the common CNN and an output unit, the learning program causing a computer to perform: a process of training the common CNN and a first CNN among the plurality of CNNs of the discriminator, using a plurality of data sets of a medical image and an image data of a first disease region in which a first disease appears in the medical image, such that information indicating the first disease region is output from the output unit of the first CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN; and a process the common CNN and a second CNN among the plurality of CNNs of training the discriminator, using a plurality of data sets of a medical image and an image data of a second disease region in which a second disease having at least one of a medical causal relationship or an anatomic causal relationship with the first disease appears in the medical image, such that information indicating the second disease region is output from the output unit of the second CNN among the plurality of CNNs in a case in which the medical image is input to the input unit of the common CNN, wherein after the discriminator being trained, the common CNN outputs a feature amount map common to a case in which the first disease is discriminated and which the second disease is discriminated.
This invention relates to a machine learning system for medical image analysis, specifically for training a discriminator network to identify disease regions in medical images. The system addresses the challenge of detecting multiple diseases with related medical or anatomical causal relationships using a shared feature extraction mechanism. The discriminator comprises a common convolutional neural network (CNN) with an input and output unit, connected to multiple specialized CNNs, each with their own input and output units. The system trains the common CNN and a first specialized CNN using medical images and corresponding first disease region data, enabling the discriminator to output information indicating the first disease region when the medical image is input. Similarly, the common CNN and a second specialized CNN are trained using medical images and second disease region data, where the second disease has a causal relationship with the first disease. After training, the common CNN generates a shared feature map that can be used to discriminate both the first and second diseases. This approach improves efficiency by reusing learned features across related disease detection tasks while maintaining accuracy. The invention is implemented as a non-transitory computer-readable medium storing the training program.
14. A non-transitory computer readable medium for storing a learning program according to claim 13 , the learning program further causing a computer to perform: a process of training the common CNN, using the plurality of data sets of a medical image and an image data of a first disease region in which the first disease appears in the medical image and the plurality of data sets of a medical image and an image data of a second disease region in which the second disease appears in the medical image, such that a feature amount data of the medical image is output from the output unit of the common CNN in a case in which the medical image is input to the input unit of the common CNN.
This invention relates to a computer-readable medium storing a learning program for training a convolutional neural network (CNN) to analyze medical images. The problem addressed is the need for an efficient and accurate method to train a CNN to detect and differentiate between multiple diseases in medical imaging data. The solution involves a common CNN architecture trained on diverse datasets to extract feature data from medical images, enabling identification of disease regions. The learning program trains the CNN using multiple datasets, each containing medical images and corresponding image data of disease regions where specific diseases appear. The training process ensures that the CNN can output feature data from its output unit when a medical image is input to its input unit. This feature data represents the presence and characteristics of disease regions within the medical image. The CNN is trained to distinguish between at least two different diseases, ensuring robust performance across varied medical imaging scenarios. The trained CNN can then be used to analyze new medical images, extracting relevant feature data for disease detection and diagnosis. This approach improves the accuracy and efficiency of medical image analysis by leveraging a shared CNN model trained on diverse datasets.
15. A discriminator that is trained by the learning program according to claim 13 .
A discriminator is a neural network component used in generative adversarial networks (GANs) to distinguish between real and generated data. The discriminator is trained by a learning program that employs a training dataset containing real data samples. The learning program iteratively adjusts the discriminator's parameters to improve its ability to accurately classify real and generated data. During training, the discriminator receives both real data from the training dataset and synthetic data produced by a generator. The discriminator outputs a probability score indicating whether the input data is real or generated. The learning program updates the discriminator's parameters based on the discriminator's performance, using techniques such as gradient descent. The training process continues until the discriminator achieves a desired level of accuracy or convergence. The trained discriminator can then be used to evaluate the quality of generated data or to refine the generator's output. This approach is particularly useful in applications like image synthesis, data augmentation, and anomaly detection, where distinguishing real from synthetic data is critical. The discriminator's training involves minimizing a loss function that measures the difference between its predictions and the true labels of the input data.
16. A disease region discrimination apparatus comprising: an image acquisition unit that acquires a medical image which is a discrimination target; and the discriminator according to claim 9 that discriminates a first disease region in the medical image which is the discrimination target.
This invention relates to medical imaging and disease region discrimination, specifically addressing the challenge of accurately identifying diseased areas within medical images. The apparatus includes an image acquisition unit that captures a medical image to be analyzed and a discriminator that processes the image to detect a first disease region. The discriminator uses a trained model to classify regions of the image, distinguishing between healthy and diseased tissue. The model is trained using a dataset of medical images with labeled disease regions, enabling it to recognize patterns associated with specific diseases. The discriminator may also incorporate additional features, such as image preprocessing steps to enhance contrast or reduce noise, and post-processing to refine the detected regions. The apparatus is designed to improve diagnostic accuracy by automating the detection of disease regions, reducing human error and increasing efficiency in medical imaging analysis. The system can be applied to various imaging modalities, including MRI, CT, and ultrasound, and is adaptable to different types of diseases, such as tumors, lesions, or other abnormalities. The invention aims to provide a reliable, automated tool for medical professionals to assist in diagnosis and treatment planning.
17. The disease region discrimination apparatus according to claim 16 , further comprising: a display control unit that displays a discrimination result of the discriminator on a display unit.
A disease region discrimination apparatus analyzes medical images to identify and classify regions of interest, such as diseased or abnormal areas, within the images. The apparatus includes an image acquisition unit that captures or receives medical images, such as X-rays, MRIs, or CT scans, from a medical imaging device. An image preprocessing unit processes these images to enhance features relevant to disease detection, such as noise reduction, contrast adjustment, or normalization. A feature extraction unit then analyzes the preprocessed images to identify key features, such as texture, shape, or intensity patterns, that may indicate the presence of disease. A discriminator, trained using machine learning or statistical models, evaluates these features to classify regions of the image as diseased or healthy. The apparatus further includes a display control unit that presents the discrimination results on a display unit, visually highlighting the identified disease regions for medical professionals. This allows for improved diagnostic accuracy and efficiency in medical imaging analysis. The system may also include a storage unit to retain processed images and results for future reference or further analysis.
18. A disease region discrimination apparatus comprising: an image acquisition unit that acquires a medical image which is a discrimination target; and the discriminator according to claim 16 that discriminates a second disease region in the medical image which is the discrimination target.
This invention relates to medical imaging and disease region discrimination, specifically addressing the challenge of accurately identifying disease regions within medical images. The apparatus includes an image acquisition unit that captures a medical image to be analyzed and a discriminator that processes the image to detect a second disease region. The discriminator uses a trained model to distinguish between healthy and diseased areas, leveraging features extracted from the image. The discriminator may employ a convolutional neural network or other machine learning techniques to enhance detection accuracy. The system is designed to improve diagnostic precision by automatically identifying disease regions, reducing reliance on manual interpretation and minimizing human error. The apparatus can be integrated into existing medical imaging systems, such as MRI, CT, or ultrasound, to provide real-time or batch analysis of medical images. The invention aims to streamline disease detection workflows, particularly in radiology, by automating the identification of abnormal regions, thereby supporting faster and more consistent diagnoses.
19. A non-transitory computer readable medium for storing a disease region discrimination program that causes a computer to perform: a process of acquiring a medical image which is a discrimination target; and a process of allowing the discriminator according to claim 9 to discriminate a first disease region in the medical image which is the discrimination target.
This invention relates to medical image analysis, specifically discriminating disease regions in medical images using machine learning. The problem addressed is the need for accurate and automated detection of disease regions in medical images, such as tumors or lesions, to assist in diagnosis and treatment planning. The invention involves a non-transitory computer-readable medium storing a disease region discrimination program. The program causes a computer to perform two main processes. First, it acquires a medical image that is the target for discrimination. This image could be from modalities like MRI, CT, or X-ray scans. Second, it uses a discriminator to identify a first disease region within the acquired medical image. The discriminator is a machine learning model trained to recognize patterns associated with specific diseases. The discriminator may use convolutional neural networks (CNNs) or other deep learning architectures to analyze pixel-level features and classify regions as diseased or healthy. The discriminator may also incorporate attention mechanisms to focus on relevant areas of the image, improving accuracy. The program may further include preprocessing steps to enhance image quality, such as noise reduction or contrast adjustment, before discrimination. The output is a marked or segmented image highlighting the detected disease region, aiding medical professionals in diagnosis. The invention aims to improve diagnostic efficiency and reduce human error in medical image interpretation.
20. A non-transitory computer readable medium for storing a disease region discrimination program that causes a computer to perform: a process of acquiring a medical image which is a discrimination target; and a process of allowing the discriminator according to claim 9 to discriminate a second disease region in the medical image which is the discrimination target.
This invention relates to medical imaging and disease region discrimination, specifically addressing the challenge of accurately identifying disease regions within medical images. The system involves a non-transitory computer-readable medium storing a disease region discrimination program designed to execute on a computer. The program performs two key processes: first, it acquires a medical image designated as the discrimination target, which may include various imaging modalities such as X-rays, MRIs, or CT scans. Second, it utilizes a discriminator to analyze the medical image and identify a second disease region within it. The discriminator is a specialized algorithm trained to recognize patterns indicative of disease, distinguishing affected areas from healthy tissue. This approach enhances diagnostic accuracy by automating the detection of disease regions, reducing human error and improving efficiency in medical analysis. The system is particularly useful in clinical settings where rapid and precise disease identification is critical for treatment planning.
Cooperative Patent Classification codes for this invention. Click any code to explore related patents in that topic.
September 26, 2019
February 8, 2022
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.